aethron-30b

surplus-launcher: 0xbd36b1714c6C140d096d91e8C7a172Bac8208046

aethron-30b is the agent model of AethronAI, a read-only onchain intelligence tool. It is Qwen/Qwen3-30B-A3B-Instruct-2507 (30.5B parameters, ~3.3B active per token, mixture of experts) fine-tuned with LoRA and merged into full bf16 weights, trained to work the way Aethron Agent must:

• call AethronAI's read-only tools (scan_token, trace_wallet, compare_tokens, explain_transaction, ...) with the right network and address,
• answer only through a final_answer tool call: short statements of kind fact, inference, unknown or reply, each citing evidence ids (E3, ...) from the tool results,
• write in the language selected in the dashboard, whatever language the question is in.

It is meant to run inside AethronAI, with its system prompt and tool definitions. Every answer is checked by AethronAI's evidence validator before it is shown; statements with figures or addresses not found in the cited evidence, safety claims, advice or identity claims are dropped.

Training

Base model	Qwen/Qwen3-30B-A3B-Instruct-2507, bf16
Method	LoRA, rank 16, alpha 32, dropout 0.05, on the attention projections q, k, v, o (experts and router stay frozen), merged into the base weights
Data	AethronAI agent transcripts, up to 8,192 tokens each, about 10% held out by address
Loss	assistant turns only (tool calls and the final answer, ~9% of tokens)
Schedule	2 epochs, learning rate 2e-4 cosine, effective batch 8

Data. Questions in English, Spanish, German, French, Portuguese, Russian, Chinese, Japanese and Korean about real tokens, wallets and transactions on Robinhood Chain, Base, BNB Smart Chain, Arc and Solana were run through the real AethronAI agent with live chain data. The teacher was an open-weights model (Qwen/Qwen3.8-27B through Hugging Face Inference Providers). A transcript was kept only if every lookup succeeded, the model answered through final_answer without a reminder, and the evidence validator rejected at most one statement (which was then removed). No Claude output was used.

The same data trained aethron-20b, a LoRA adapter for gpt-oss-20b.

Limitations

• A small first version: a few hundred training transcripts. It can lead with secondary details (for example transfer activity before balances).
• Trained on AethronAI's tool schema and system prompt; it is not a general chat model.
• Onchain data can be incomplete or wrong, and the model can misread it. Answers are not financial advice.

Use

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("aethronai/aethron-30b")
model = AutoModelForCausalLM.from_pretrained("aethronai/aethron-30b", dtype=torch.bfloat16, device_map="auto")

Prompts use the Qwen3 chat template with AethronAI's tools (tokenizer.apply_chat_template(messages, tools=tools, add_generation_prompt=True)). vLLM serves it like the base model, with the hermes tool-call parser.